Jeong DEA SE generation

suppressPackageStartupMessages({
  library(SummarizedExperiment)
  library(SEtools)
  library(ggplot2)
  library(plgINS)
  library(edgeR)
})

Loading

# load splice SE
se <- readRDS("data/raw/jeong.raw.SE.rds")

Prep

# modify colData
se$miRNA <- factor(c("DKO","WT","WT","DKO"))
se$rep <- factor(1:2)

# filter raw data
filter <- c(10,2)
se <- se[which(rowSums(assay(se) >= filter[1]) >= filter[2]),]

logAssays <- function(se, control){
  ## SE object with logCPM & logFC assays
  assays(se)$logcpm <- log1p(cpm(calcNormFactors(DGEList(assay(se)))))
  se <- SEtools::log2FC(se, controls = se$miRNA==control, fromAssay = "logcpm")
  return(se)
}

se <- logAssays(se, "WT")

PCA

# plot PCA to check for batch effect

## combined
plgINS::plPCA(assays(se)$logcpm, as.data.frame(colData(se)), colorBy = "rep",
              add.labels = FALSE, annotation_columns = colnames(colData(se))[3])

DEA

# define variables for individual DEAs
ctrl <- "WT"
mirna <- unique(as.character(se$miRNA[se$miRNA!=ctrl]))
dea.names <- paste0(unique(se$miRNA[se$miRNA!=ctrl]),"vWT")
# select reference factors
se$miRNA <- relevel(droplevels(se$miRNA), ref=ctrl)

source("functions/dea.R")

# DEA over all treatments
se <- DEA(se, name="all", model = ~rep+miRNA, model0 = ~rep)

source("functions/dea.R")

# DEAs of individual treatments
for(i in 1:length(mirna)){
  se <- DEA(se, use=se$miRNA %in% c(ctrl,mirna[i]), name=dea.names[i], model = ~rep+miRNA, model0 = ~rep)
}

Heatmap

# heatmap function

makeHM <- function(se, sig, nr=20, logcpm=FALSE){
  for(i in sig){
    if(sum(rowData(se)$DEA.all$FDR < i) > nr){
      cat("FDR <", i, "\n")
      # logFC heatmap
      SEtools::sehm(se[,order(colData(se)$miRNA)], row.names(se)[rowData(se)$DEA.all$FDR < i], gaps_at = "readtype",
                breaks=TRUE, do.scale = FALSE, show_colnames = FALSE, assayName = "log2FC", anno_columns = "miRNA")
      # logcpm heatmap
      if(logcpm){
        SEtools::sehm(se[,order(colData(se)$miRNA)], row.names(se)[rowData(se)$DEA.all$FDR < i], gaps_at = "readtype",
                breaks=TRUE, do.scale = TRUE, show_colnames = FALSE, assayName = "logcpm", anno_columns = "miRNA")
      }
      break
    }
  }
}

# allocation
sig.lvl <- c(1e-5,1e-3,.05,.1,.5,.8)

# heatmap of DEA.all
makeHM(se, sig.lvl, logcpm = TRUE)

## FDR < 1e-05

Export

# round
source("functions/roundSE.R")
se <- roundSE(se)
# export
saveRDS(se, "data/jeong.DEA.SE.rds")

# import
se <- readRDS("data/jeong.DEA.SE.rds")

Stats

# number of significant transcripts
sapply(rowData(se)[,grepl("DEA",colnames(rowData(se)))], function(x) sum(x$FDR < .05) )

##    DEA.all DEA.DKOvWT 
##       6473       6473

up-/downregulations at different significance levels

# check number of positive & negative logFC for different significances

sigsDF <- function(se, sig, dea, thr){
  data.frame(sigLevel=rep(sig,3),
             counts=c(sapply(sig, function(x) sum(sign(assays(se[rowData(se)[[dea]]$FDR < x,])$log2FC) == -1 & 
                                                    abs(assays(se[rowData(se)[[dea]]$FDR < x,])$log2FC) >= thr)),
                      sapply(sig, function(x) sum(sign(assays(se[rowData(se)[[dea]]$FDR < x,])$log2FC) == 0 & 
                                                    abs(assays(se[rowData(se)[[dea]]$FDR < x,])$log2FC) >= thr)),
                      sapply(sig, function(x) sum(sign(assays(se[rowData(se)[[dea]]$FDR < x,])$log2FC) == 1 & 
                                                    abs(assays(se[rowData(se)[[dea]]$FDR < x,])$log2FC) >= thr))
                      ),
             sign=c(rep("-1",length(sig)),
                    rep("0",length(sig)),
                    rep("1",length(sig))
                    ),
             dea=rep(dea,length(sig)*3)
             )
}

# for each model & DEA: find number of down- & upregulations at different significance levels

## significance levels of interest
sig <- c(1e-10, 1e-5, .05, .1, .5, .8)
## only absolute log2FC greater than this will be considered
fc.thr <- .5

## create dataframes with count informations
### [model = ~miRNA, model0 = ~1]
sigs <- lapply(grep("DEA",colnames(rowData(se)),value=TRUE), function(x){
  sigsDF(se, sig, x, fc.thr)
})
sigs <- data.frame(do.call("rbind",sigs))

### [model = ~miRNA, model0 = ~1]
ggplot(sigs, aes(x=as.factor(sigLevel), y=counts, fill=sign)) + geom_bar(position="dodge", stat="identity") + facet_wrap(~dea)